Existing methods for dynamic obstacle detection and tracking have lots of shortages, such as the high false detection rate for geometric characteristics based methods, and severe influences of scanning angle and/or distance on features and motion based methods. All these issues make existing methods can't satisfy the requirements of real traffic scenarios. Targeting at aforementioned disadvantages, two methods are proposed: a multi-feature fusion based method for dynamic obstacle detection and tracking, and a spatial-temporal characteristic vector based method for obstacle recognition. Firstly, the echo pulse width of obstacles is considered based on the geometric features of dynamic obstacles to increase the accuracy of detection and tracking results. Secondly, a spatial-temporal feature vector is constructed with considering time dimension and spatial dimension information of obstacles. Thus, the SVM (support vector machine) method is used to identify dynamic obstacles to improve the recognition accuracy. Finally, the accuracy and validity of the proposed methods are verified by an autonomous vehicle in real traffic environments.